The Challenge
A leading business group and benchmark in the market for the distribution of steel products in Latin America, is constantly innovating in all areas involved in its commercial operations.
Shortly after having most of its processes digitized, the need to expand automation to include those processes that still had employees for data entry arrived in the systems area. A slow process, very prone to failures and not scalable.
Initially, the systems area ventured into OCR technology and data extraction from templates, however, they quickly came to the conclusion that this configuration process was very expensive and largely dependent on the development area.
After some survey meetings, the Mototech team had the necessary information to help the client accurately determine specific objectives and possible metrics to quantify their success.
The purchasing and logistics sectors require daily entry into the ERP of a large amount of data from invoices and purchase receipts, as well as from sales receipts. In addition, to complete the process, the digitized document must be entered in the SGD (Document Management System) correctly classified.
The goal is to stop wasting time, money and resources on manual data entry, and for the implemented solution to be scalable.
Our Solution
The most accurate, fastest and most cost-effective solution is to use artificial intelligence to capture cognitive data. This technology, unlike humans, have no distractions and do not need any rest.
Mototech’s team of data scientists use a sufficiently representative sample of the document repository and apply deep learning algorithms to train neural networks capable of understanding the structure and semantics of the information contained in each type of document, with the aim of recognize and extract some specific data that we call entities.
The development team implements both the integration with the data source, that is, the way in which the documents enter the cognitive process, as well as the integration with the business, that is, the output of the inference process integrated with existing processes and systems. -existing.
Results
The documents enter through an email that centralizes the documents. An automated process takes care of downloading them and running the inference process against the AI model. The solution contemplates the execution of data validations, and if approved, inserts the extracted data in the company’s ERP. While if any of the validations fail, an operator is informed to make manual adjustments in an interface designed for this purpose.
The cognitive process performs the automatic upload of documents to the Document Management System classified by supplier and with their respective unique voucher identifier.
AI-powered data capture helps improve time management, accuracy, and productivity and is an ideal solution for businesses with high document variability.
Are you interested in increasing the efficiency of your processes? Tell us your business case and we will gladly design a solution adapted to your needs.
Technologies
Algorithms: Transformers, Embeddings, TL-GAN, GAN-based noise, YOLO, Faster R-CNN, NER (Named entity recognition), LSTM
Libraries: Tensorflow, AsanteOCR, Camelot, QR detection, OpenCV, spaCy
Development: Stack MEAN (Mongo, Express, Angular, Node), FastAPI